De la Ingenier

1
De la Ingeniería Inversa a la Simulación en
Centros de Computación de Alto Desempeño

• Prof. Pierre Boulanger
• University of Alberta
• Dept. Computing Science

2
Canada
3
Rocky Mountains
4
Edmonton, Alberta
5
University of Alberta
The University of Alberta is home to over 2,500
international students who come each year to
study in Edmonton, Alberta.  Founded in 1908, the
University of Alberta has more than 36,000
students and offers more than 200 undergraduate
and 170 graduates programs in 18 academic
faculties.
6
Why use Rapid Product Development?

• There are many business pressures on industry
  today affecting new products
• A demand for reduced lead times
• Increased product variety and quality
  requirements
• Worldwide competition
• Short product life cycles
• Smaller product runs and customized products
• Just-in-time manufacturing

7
Enabling Technologies for RPD

• Computer Aided Engineering (CAE)
• Rapid Prototyping Machines
• Conversion Technologies
• Reverse Engineering
• Dimensional Validation
• 3-D Measuring Devices
• Virtual Display Systems
• Production Simulation Tools
• Advanced Materials

8
Toward an Integrated RPD Prototyping Environment
9
Digitizing the World

• Mechanical/Triangulation Systems
• Photogrammetric/Triangulation Systems
• Industrial Computer Tomography

10
Many Sensors for the Job
11
AMMI Lab 3D Color Digitizer
Color Camera
Range Sensor
Touch Probe
50 Micron Precision Scan Rate of 3500 pts/sec
12
Basic Optical Triangulation

• Sine-law 1 distance2 angles
• Basic optical triangulation pin-hole model
• The light beam generated by the laser/light
  projector is deflected at an angle a
• The position of the diffused image of the
  laser/light beam is measured by the imaging
  device
• The 3-D coordinate of the light point on the
  object is calculated
• The range error on the measurement is

13
NRC/VIT Color Range Sensor
14
Photogrammetric/Laser Scanner
HandyScan from Creaform 3D
15
Final Scanned Model
16
Industrial CT Scanner
Allow to Scan Inner And Outer Structures Scanning
Limited by Material Properties
17
3D Intelligent and Configurable System for
Deformable Parts Inspection

• Develop an intelligent system based on optical
  range measurements that accelerates the quality
  control process of 3D deformable parts
• Faster, less expensive and online means to
  control the manufacture and assembly of
  deformable parts such as those constructed of
  sheet metal, composites and plastics
• Precarn, Creaform, UofA, Laval U., and EAFIT

18
What is Inspection?
19
Motivation for New Framework

• Current inspection systems have been developed
  for rigid parts
• The growing usage of deformable materials have
  brought new challenges for manufacturing
  industries
• The shape of deformable parts will vary unless
  they are precisely constrained in their final
  positions. This makes inspection a tedious and
  inefficient operation.

20
Current Deformable Part Inspection Pipeline
21
Inspection Process in the New Framework
22
How to Deal With Deformations?

• Two deformation models are currently under
  investigation
• Polygon Morphing
• Large Deformation Shell Model Using FEM

23
CAD Model to Surface Mesh
24
Deformation Animation d 0
25
Deformation Animation d 0.1
26
Deformation Animation d 0.2
27
Deformation Animation d 0.3
28
Deformation Animation d 0.4
29
Deformation Animation d 0.5
30
Deformation Animation d 0.6
31
Deformation Animation d 0.7
32
Deformation Animation d 0.8
33
Deformation Animation d 0.9
34
Deformation Animation d 1.0
35
Inspection Results Before and After
36
Landmark Based Polygon Morphing

• Pros
• Easy to implement
• Can model local as well as global deformations
• Can deal with large deformations
• Cons
• Do not represent very well real materials
• Hard to define deformation tolerances
• Hard to relate to real physical properties
• Polygon morphing only guarantees that the
  morphing of the prototype mesh is accurate near
  the landmark points.

37
Large Displacement Formulation (LDF) of Shell
Elements

• We need a deformation method that is more
  physically based to improve accuracy
• We are developing in collaboration with EAFIT a
  physical model for a Large Displacement
  Formulation (LDF) shell elements solved using
  FEM.

38
Large Displacement Formulation (LDF) Using Shell
Elements
39
Reverse Engineering Work Flow
40
Production of a Hand-Made Toy in Two Weeks
41
Museum Replicas
42
Reverse Engineering of a Broken Watch
43
Watch Bracelet Reconstruction
44
Rapid Virtual Prototyping

• Once a 3D model is created, virtual prototyping
  allows product testing without the need to build
  real prototype
• Allows for shape and functional optimization
• Allow to tract the complete life cycle of a
  product
• Rapid Virtual Prototyping require powerful
  computing infrastructure especially if it is
  interactive

45
Why is this Watch Bracelet Break?
46
How Can We Improve the Design of a Crank Shaft?
47
Why Is This Ford Truck Cutch Fork Break and How
Can We Avoid This?
48
High Stress Point Due to Miss-Alignment
49
The UofA/EAFIT Virtual Wind Tunnel
50
UofA/EAFIT Virtual Wind Tunnel Architecture
51
Terrain Model of Mount-Saint Helens
Terrain Model After Compression and Hole Filling
Terrain Model Rendering
52
From Terrain Model to CFD Mesh
53
Low Altitude Airflow Over Mount Saint-Helens
54
Buoyancy Model Over Medellin
Low Altitude Air Flow
Fusion of STRM NASA Data and Landsat infrared
images
55
CFD Simulation of Francis Turbine Project
56
Interactive CFD User needs
57
First Level of Function Segregation
58
Multi-Modal Interface for CFD
59
Type of Modalities

• Modalities used for the interface
• Visual Mono/Stereo/CAVE
• Haptic
• Perception of fluids flow
• Objects manipulations
• Setting of boundary conditions
• Sonification of fluids

60
Collaborative Visualization/Simulation Using AG,
VNC, and Simulation Server
61
Collaborative Exploration and Steering for CFD
Using Renata/Canarie/Géant
EAFIT/ Medellin
Los Andes/Bogota
High-Speed Network Canarie/Renata/Géant
AMMI Lab/ Edmonton
LIMSI/Paris
62
West Grid is a Large Resource of Compute Power
63
Past Compute Power for VWT

• Arcturus
• SGI Origin model 3900
• 256 processors (700 MHz IP35)
• CPU MIPS R16000 Processor
• FPU MIPS R16010 Floating Point Unit
• Main memory size 262144 Mbytes
• Instruction cache size 32 Kbytes
• Data cache size 32 Kbytes
• Secondary unified instruction/data cache size
  8 Mbytes
• Fabric RAID

64
New Compute Power for VWT

• 64 Itanium CPUs Connected by a 35 Gb/s NUMA
  bus.
• 6 ATI graphics cards
• 256 GB of memory
• 5 TB Disks
• 2 FPGA Nodes
• 10 Gb/s networking

65
Next Step GPGPU Implementation of CFD Equations
66
Governing Equations
67
Data Flow Between CPU and GPU
68
GPU Flexible and Precise

• Modern GPUs are deeply programmable
• Programmable pixel, vertex, video engines
• Solidifying high-level language support
• Modern GPUs support high precision
• 32 bit floating point throughout the pipeline
• High enough for many (not all) applications

69
UofA Simulation Server

• The project objectives are
• Allow Truly distributed Simulation and
  Visualization
• Allows to separate simulation time from real-time
  visualization requirements
• Allow multi-users to interact with the simulator
• Allow real-time modifications of boundary
  conditions and simulation parameters

70
Visualization vs Simulation Software Architecture
71
Conclusion

• Our ambitious goal of creating a true real-time
  virtual wind tunnel is getting closer.
• The new Altrix 4700 with six GPUs and two FPGA
  nodes may help us break the Teraflops barrier
• Our collaboration with EAFIT have been very
  fruitful and we intend to expand this
  collaboration further.